International Journal on Natural Language Computing (IJNLC) Vol. 5, No.

2, April 2016

DEVELOPMENT OF LEXICONS GENERATION TOOLS

FOR ARABIC: CASE OF AN OPEN SOURCE
CONJUGATOR
Mourchid Mohamed1, El Faddouli Nour-eddine2 and Amali Said3
1

MIC search team, Laboratory MISC, Faculty of sciences, Ibn Tofail University KenitraMorocco
2
RIME search team, LRIE laboratory, Mahammadia School of Engineers, Mohammed
Vth University in Rabat, Morocco
3
OMEGA search team, Laboratory LERES, Faculty FSJES, Moulay Ismal University
Meknes-Morocco

ABSTRACT
The dictionary resources are very important for Natural Language Processing (NLP). Generating high
quality dictionary resources is a crucial step for the success and effectiveness of NLP application.
Linguistic information about lexical database is complex, large size and various (ie, phonological,
morphological, syntactic, semantic and pragmatic). Among such lexical database entries, we find
conjugated verbs. To this end, we present in this paper the open source mobile application of our
conjugator that we developed in Java platform under the Android. This Conjugator allowed us to generate
a lexicon of more than 18667 conjugated verbs. This lexicon will be used to generate textual words. The
resultant lexicon can be used in various applications such as morphological analysis (lexical approach),
text indexing, etc

KEYWORDS
NLP, Lexical databases, conjugator, Root, Pattern, textual words, morphological analysis.

1. INTRODUCTION
In the Arabic language, each word having a meaning consists of a root and a pattern. So we can
represent all the Arabic words by a matrix in which the patterns are the columns and roots are the
rows. A word is simply an element in this matrix [1] [3] [5] [9] (see Figure. 1).

Figure 1. Matrix Root/Pattern

DOI: 10.5121/ijnlc.2016.5202

13

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

With this matrix, we can perform two operations [7][10]:

- Analysis: extract the pattern and the root of a given word.
- Generation: building a word from a column (pattern) and a row (root).
Figure 2 shows an example of using this matrix to extract the root and the pattern of the word
(MaKtaba , Library) and to generate the word from the root
(Root KTB,
Write)and the pattern .

Figure 2. Operations: Analysis and Generation

In this paper, we present the approach of our conjugator developed in Java in two versions. The
first is a mobile application under android, while the second is a desktop application to power a
lexicon for morphological analysis.
The principle of morphological generation is presented in the second section.
In the third section, we describe the different approach steps of our conjugator and the used
repository.
The fourth section is dedicated to description of the conjugator mobile application.
The exploitation of conjugator to enrich the lexicon of morphological analysis by generating
textual words is described in the fifth section.
In conclusion, we discuss the results and the main perspectives of our research.

2. MORPHOLOGICAL GENERATION PRINCIPLE

Morphological generation is a succession of morphological operations applied to an initial word
accompanied by a set of attributes, to produce a final form of the word.
These attributes are for :

Names: Number (singular, dual, plural); case (nominative, accusative, gerund);

determination (defined, undefined), annexation (annexed)

14

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

Verbs: Aspect (perfect, imperfect, imperative); voice (passive, active); Number (singular,
dual, plural); gender (male, female); person (first person, second person, third person);
case ( nominative, accusative, apocope ); insistence.

2.1. Generation Methods

There are three generation methods differentiated by their approaches [5].
2.1.1. Successive Transformations Method
This method relies on applying transformations progressively until obtaining of the final form.
These transformations are shown as follows:
Initial word + AT1====> word1
word1+ AT2 ====> word2
.
wordn-1 + ATn ====> final word
Where ATi is an attribute.
The figure 3 shows how to get final word ( the sciences) from the initial word
using the set of attributes(Plural , Defined , Accusative).

(science )

Figure 3. The final word is ( AL-OLOMA, the sciences)

2.1.2. Method of Pre-established Models
This method is based on the use of a set of predetermined models each of which is associated
with a class of words having the same characteristics.
Thereby the word generation is done in two steps:
Determination of suitable model from the characteristics of the original word.
Performing a substitution operation from the initial word and the corresponding model.
The figure 4 depicts the steps followed to obtain the mould from a set attributes.
15

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

Figure 4. Obtaining the mould

2.1.3. MIXED METHOD

We apply first the pre-established model method. Then we apply the successive transformations
method on the result of the first step (See figure 5).

Figure 5. The final word is ( YAQoRaAu, They read)

3. CONJUGATOR OF VERBS [5]

The method adopted for conjugation of the verb is the mixed method.
Thus, the conjugation operation of a verb proceeds in six steps:
16

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

Determining the verb class (see Figure. 6) by consulting the lexicon t of triliteral verbs or
make a treatment based on the length of the verb and the location of certain consonants.

Figure 6. Determining of the verb class

The adequate model is determined from the triple (class tense, time, voice). (see Figure
7).

Figure 7. Determining the model.

The triple (person, number, gender) and the model number allow to extract the desired
mould (see Figure 8).

17

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

Figure 8. Determination of the mould

Substitution operation consist to replace the mould numbers with the corresponding
consonants of the verb as shown in the example of table 1
Table 1. The substitution operation.

Verb

Mould

Substitution

( DaKaLa, Enter)

123

" (YaDoKuLu,Enter)

Application of the transformation rules (T.R)

insistence) of imperfect tense.

to deal the case (Apocope, accusative,

For each case, we define the actions to perform on the consonants and vowels of the verb. Table 2
shows an example of the transformations actions:
Table 2. Some transformations actions.

Action
number

Action

Coding of actions

Change the last character of the vowel part by " "

remove the last character of the consonant part

remove the last character of the vowel part

Add the character ' 'to the last position

Table 3 is an example of application of the transformation rules.

18

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016
Table 3. Application of the RT.

Rule number

Transformations actions

Example ( Before)

Example ( After)

2,3

2,3,4

Spellchecking: This step involves performing the possible corrections on the verb, based
on the lexicon of the corresponding anomalies and corrections; an example is giv-en in
table 4.
Table 4. Anomaly and correction.

Verb anomaly

Verb after correcting

&

&

4. IMPLEMENTATION AND TESTING OF OUR CONJUGATOR

We realised our conjuguator based on the rules outlined in the previous sections. It consists of
four modules (see Figure 9).

Figure 9. Conjugator modules

The presentation module enables communication with the user, which can be a person or
an application in which our conjugator can be integrated. This module makes the
necessary checks on the verb to conjugate and determines the pattern to apply.
The substitution and transformation module enables to apply the transformation rules on
the pattern determined by the previous module.
The correction module applies correction rules on the result provided by the
transformation module to obtain the final shape.
The acquisition module and control will be used by a linguist to feed the repository by the
verbs, the models and the correction rules missing.
19

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

The repository of our conjugutor contains all relevant data: verbs, models and correction rules.
These data can be stored in several forms (relational DB, XML, Jason ...).
We implemented our conjugator, in its first version, as an Android mobile application using the
Java language for coding, the DBMS SQLite for managing the repository and XML to generate
mobile interfaces. The mobile application is open source and can be downloaded from the RIME
search team web site ( http://rime.emi.ac.ma/arabic_conjugator/conjugator.apk).
Figure 10 shows the interface that allows the user to enter the verb, choose the aspect, the voice,
and the mode of conjugation.

Figure 10. Data entry

These data will be recovered by the presentation module that will determine the model to apply.
The transformation rules for the latter will be applied by the substitution module whose result will
be processed by the correction module. The final result obtained will be communicated to the user
(see Figure 11).

20

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

Figure 11. Conjugation of the verb '

' at the accomplied tense

The current repository of our conjugator contains:

Table 5. Statistics.

Arabic term
Number
Roots
6413
Triliteral roots
5635
Roots quadrilitaires
592
Canonical schemas
199
Irreducible trilitaires verbs () *
6138
Reducible trilitaires verbs ( "+ )
11832
Irreducible quadrilitaires verbs ( ) * 452
Reducible quadrilitaire verbs ( "+ )
245

5. GENERATION OF TEXTUAL WORDS

In the previous section, we used our conjugator as core of a mobile application.
In this section, we use our conjugator to generate textual words using the lexicon of verbs noted
M0. These textual words will enrich the dictionary that can be exploited for morphological
analysis.

5.1. GENERATION OF INFLECTED FORMS

The Arabic words are built or bending. The words built have an invariant form whatever their
position in the text; this is the case of tools words (prepositions, conjunctions, etc...).

21

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

The bending words have a termination that changes depending on the case: nominative,
accusative, apocope, etc
We try to generate the inflected forms (lexical kernel of the textual word), noted M0C, for each
verb in the lexicon M0. The principle of generation processing of inflected forms is as follows:
M0 is the set of verbs
M0 = v1, v2,...
For each m in M0
if m.VALG = imperfect then
+ Inflected_Form1 Generate_Accusative(m)
+ Inflected_Form2 Generate_ Apocope (m)
+ Add(Inflected_Form1, M0C) {to add Inflected_Form1 to the lexicon M0C}
+ Add(Inflected_Form2, M0C) {to add Inflected_Form1 to the lexicon M0C}
End if
End for
Where: VALG = Grammatical value of the verb (perfect, imperfect, imperative)

5.2. GENERATION OF TEXTUAL WORDS

The Lexicons M0, M0C represent respectively the lexical words and textual words not having
prefixes and suffixes. Arabic prefixes are sets of letters attached to the beginning of the lexical
word and written as part of it ( for example, prepositions
Bi with , , Li, for,
Ka, like and conjunctions Wa, and , Fa, and, the determiner ',' '| aL', 'the ) ,
while suffixes are sets of letters and pronouns attached to the end of the word and written as part
of ( for example, the object pronoun hum , them) [6].
5.2.1. CONCEPT OF MICRO-SYNTAX [2][4] [8]
We have defined a micro-syntax that represents the set of rules for arranging correctly the
attachment of prefixes and suffixes to a word of the lexicons M0 or M0C to generate a textual
word.
Formally, we can write: Textual word =
+ FS +
with:
Pi: is a prefix.
FS: is a simple form that can be a lexical word owned at M0 or M0C.
Si: is a suffix.
The attachment of prefixes and suffixes to words presents four major problems:
1) The sequences of prefixes (
) and suffixes (
) are not arbitrary.
Example: the sequence 1( WaBi|aL, and with the ) is valid while 1
(BiWa|aL) is not.
2) Some prefixes or suffixes can be attached only to specified simple forms; for example,
the prefix "( "Sa, will) can be attached to verbal forms at the imperfect, while the prefix
" |( "aL, the) can be only attached to nominal forms.
3) Some prefixes cannot be glued to a given word, at the same time as certain suffix; for
example, we can say "" | ( "aLWaLaDu, the boy ) or " ( " "WaLaDuHu, his son ) but
we cant say "| (" "aLWaLaDuHu).
22

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

4) The attachement of certain prefixes or suffixes to a FS is not limited to a simple

concatenation, it implies in many cases a change in the FS termination.
Example:

( his meaning)

+(

meaning)

5.2.2. GENERATION ALGORITHM OF TEXTUAL WORDS

Let P be the set of simple prefixes and compounds: P = {P1, P2 ...}.
The constitution of these sequences is done manually from any particles that exist in Arabic
(preposition, conjunctions ...).
Analogously to the prefixes, we build the set of suffixes: S= {S1, S2 }.
We define valid combinations between those elements of P and S. It constitutes thus the Pi
couples, Sj accompanied by a motion to indicate for which categories of words could be apply
(name, verb, etc. ...).
This gives what we call a text morphological generation rule (TMR):
Pi , Sj , Motion
Motion which may be of type:
Noun: all word of type noun, adjective are compatible with this Pi couple Sj.
Verb: perfective, imperfective or imperative are compatible with this Pi couple Sj.
Verb imperfective: Only words with the unaccomplished are compatible with this Pi couple Sj.
Examples :
( WaSa, and will) , (Hum, them) , imperfective
, (Wabi, and with) , (Him, them) , Noun
In practice, items are handled by classes micro-syntactic. For example, the two previous
generation rules will be represented by:
[Wabi] , [Him] , Noun
[WaSa] , [Hum] , imperfect
The proposed generation algorithm of textual words composing lexicon noted M1 is:
M=M0 M0C is the set of words to process
Let R be the lexicon rules of textual morphological generation: R= {r1,r2,.}
For each m in M
For each r in R
textual_word Generate_textual_word(m, r) { apply the rule r to the word m }
Add(textual_word, M1) {to add textual_word to the lexicon M1}
End for
End for

23

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

6. CONCLUSIONS
In this paper, we presented the architecture of a system of conjugation of Arab verbs. It operates
according to a five-step process: determining the class of verb, determining the model, the
substitution operation, applying transformation rules, and spells correcting. The results of the
performed tests are very satisfactory.
We used the conjugator to enrich the lexical database by the verbs for morphological analysis
using the dictionary-based approach. We proposed also a micro-syntax and an algorithm to
generate the textual words to enrich the same lexical database.
We intend to use our conjugator as the core of a learning environment for Arabic and especially

conjugation for kids and non-native speakers.

To complete the necessary dictionary for morphological analysis, we intend to generate the
textual words for nouns.

REFERENCES
[1]
[2]

Ali Nabil (1988) Arabic Language and Computer, [in Arabic], Taareeb.
Hlal Yahya, (1979) Learning Methods for Morphosyntactic Analysis (Experienced in the case of
Arabic and French), thesis doctoral degree, University Paris.
[3] Hlal Yahya, (1987) Generation from the root and pattern, Conference on the progress of linguistics
in Arab countries.
[4] Hlal Yahya, (1990) Morphology and syntax of the Arabic language, In Proceedings of the Arab
School of Science and Technology Applied Arabic Linguistics for Informatics (pp.201).
[5] Mourchid Mohamed, (1999) Generation Morphological and Applications, Specialty thesis of 3rd
round, Mohammed V University in Rabat-Morocco.
[6] Nizar Y. Habash, (2010) Introduction to Arabic Natural Language Processing, Morgan & Claypool
Publishers series.
[7] Abdelhadi Soudi, Antal van den BoschNizar, & Gunter Neumann (2007) Arabic Computational
Morphology, Knowledge-based and Empirical Methods, Springer
[8] Joseph Dichy, Ali Farghaly, (2003) Roots & Patterns vs. Stems plus Grammar-Lexis Specifications:
on what basis should a multilingual lexical database centred on Arabic be built? Workshop on
Machine Translation for Semitic Languages: issues and approaches New Orleans, USA.
[9] Riyad Al-Shalabi, (2005) Pattern-based Stemmer for Finding Arabic Roots, Information
Technology Journal, 4(1): p. 38-43.
[10] A.Yousfi, (2010) The morphological analysis of Arabic verbs by using the surface patterns, IJCSI
International Journal of Computer Science Issues,7(3(11)): p. 33-36.

24

International Journal on Natural Language Computing (IJNLC) Vol. 5, No.2, April 2016

AUTHORS
M. MOURCHID Doctorate degree in Computer Science in 1999; Assistant Professor
at the Computer Science Department at the Faculty of sciences, Ibn Tofail University
in Kenitra Morocco; Ongoing research interests: Natural Language Processing, Web
Semantic, and information systems.
N. El Faddouli Doctorate degree in Computer Science in 1999; Assistant Professor at
the Computer Science Department at the Mohammadia School of Engineers (EMI); 15
recent publications papers between 2007 and 2015; Ongoing research interests: elearning, Big Da ta, Natural Language Processing, and information systems. Language
Processing, and information systems

S. AMALI Doctorate degree in Computer Science in 1999; Assistant Professor at the

Computer Science Department at the Moulay Ismal University; Ongoing research
interests: e-learning, Natural Language Processing, and information systems.

25